Kimi K3 scores 32% on ExploitBench, far behind U.S. frontier models
The Decoder · rss · 2026-07-24
A report from The Decoder says Moonshot AI's Kimi K3 performed far worse than frontier U.S. models on offensive cyber tasks.
According to the British AI Security Institute and the U.S. Center for AI Standards and Innovation:
- 32% on ExploitBench for Kimi K3
- 76% for leading U.S. models
- safeguards did not reliably block exploit development or simulated attacks
The article argues the gap between Kimi K3's strong general benchmarks and weak cyber performance may align with allegations that Moonshot AI distilled Anthropic models.
More from Models
- A poster says models are still toys, but close to becoming useful — MoonL88537 · 2026-07-24
- BTL-3 trends on Hugging Face as a text-generation model for coding and tool use — badtheorylabs · 2026-07-24
- Tencent merges Hunyuan multimodal and LLM teams under Shunyu Yao — jiqizhixin · 2026-07-24
- Kimi paid upgrade opens for a fresh US account, EU users still waitlisted — evilsocket · 2026-07-24
- Laguna S 2.1 users debate whether low-precision quantization is hurting reasoning — burritoresearch · 2026-07-24
- Opus 5 is a make-or-break release for Anthropic, says antirez — antirez · 2026-07-24